|Category Reference for Presentations|
|AHE||Applied Health Economics||DEC||Decision Psychology and Shared Decision Making|
|HSP||Health Services, and Policy Research||MET||Quantitative Methods and Theoretical Developments|
* Candidate for the Lee B. Lusted Student Prize Competition
To examine tradeoffs between cervical cancer screening test performance (sensitivity, specificity) and programmatic considerations (population coverage, follow-up of screen-positive women) in low-resource settings.
Using an individual-based Monte Carlo simulation model of the natural history of human papillomavirus (HPV) and cervical cancer calibrated to epidemiologic data from Uganda, we assumed screening occurred once in a woman's lifetime at age 35 with two-visit HPV DNA testing or one-visit visual inspection with acetic acid (VIA). Model outcomes included reduction in lifetime risk of cancer and incremental cost-effectiveness ratios (ICERs). For each screening modality, we performed one- and two-way sensitivity analyses to assess the tradeoffs between 1) test sensitivity and specificity; 2) test sensitivity and screening coverage; and 3) screening coverage and follow-up rates of screen-positive women. Sensitivity, specificity and coverage were varied 30-100% and loss-to-follow-up rates were varied from 15-60% per clinical contact. To assess the tradeoff between test sensitivity and loss-to-follow-up given uncertainty, we compared HPV DNA testing with 80-100% sensitivity to VIA with 40-60% sensitivity, as loss-to-follow-up varied from 15-60%.
For both screening modalities, improving test sensitivity had greater potential to reduce the ICER of onetime screening relative to comparable improvements in test specificity, due to the potential for increased benefits relative to costs incurred. While there were similar reductions in cancer risk for comparable changes in coverage and sensitivity, the costs associated with increasing coverage were proportional to benefits, yielding stable ICERs; as sensitivity increased from 30-100%, ICERs decreased by 65% (HPV testing). A comparison of HPV DNA testing and VIA found that when loss-to-follow-up per clinical contact was low (i.e., 15%), HPV testing yielded greater cancer risk reductions than VIA at all assumed levels of sensitivity. When loss-to-follow-up rates reached 60% per visit, a perfectly sensitive HPV test yielded similar cancer risk reductions as the least sensitive VIA program considered. The rank ordering of strategies changed dramatically as sensitivity and loss-to-follow-up varied.
Available screening modalities pose tradeoffs between test performance, coverage, and follow-up. Where screening is limited, test sensitivity and follow-up of screen-positive women are critical determinants of the relative effectiveness and cost-effectiveness of prevention programs.
Compared to cytology, HPV testing has the potential to improve the effectiveness by reducing cervical cancer incidence due to improved early detection and treatment but also a higher risk of over-diagnosis and overtreatment of irrelevant lesions. We systematically evaluated benefits and harms of different HPV-based primary cervical cancer screening strategies in the German health care context.
A previously validated and published Markov model1 for the German health care context was used to analyze the trade-off between benefits and harms of different screening strategies differing by length of screening interval and test algorithms including HPV testing alone or in combination with cytology or with cytological triage for HPV-positive women. We used published German clinical and epidemiological data as well as test accuracy data from international meta-analyses. We used a benefit-harm frontier for reduction in cervical cancer cases (CC) vs. unnecessary treatment (defined as invasive therapy of < CIN 3) to visualize dominated strategies and incremental harm-benefit ratios.
Overall, HPV-based screening was more effective than cytology alone, with a relative reduction in cervical cancer incidence of 49%-90% compared to 33% - 80% with cytology alone (depending on screening intervals). The incremental gain in effectiveness with HPV screening compared to cytology was higher and incremental increase in harms was lower with extended screening intervals. Out of 18 strategies, 12 were dominated based on the benefit harm-frontier. Compared to annual cytology, which is currently the recommended standard in Germany, biennial HPV screening was similarly effective but reduced unnecessary treatment (depending on test and follow-up algorithm). In contrast, annual HPV primary screening compared to annual cytology alone would result in an incremental harm-benefit ratio of 12 - 117 unnecessary treatments per additional prevented cervical cancer case (depending on screening adherence rate).
Based on our decision-analytic benefit-harm frontier analyses, HPV-based cervical cancer screening is more effective than cytology alone, but has a higher risk of overtreatment when used in annual screening. In the German health care context, depending on screening adherence rate biennial or triennial HPV screening for women aged 30 yrs and older is similarly or more effective as annual cytology alone, but with significantly reduced unnecessary treatments.
1. Sroczynski et al. Eur J Cancer 2011;47(11):1633-1646.
Method: We used data from an independent UK clinical trial where 50 readers each read 180 mammograms - 60 with cancer and 120 normal. We selected four groups of complementary reader pairs in which a member of a Group A, expected to be more effective, is paired with a member of a Group B, expected to be less effective. The complementary AB groups are: (1) high and low experience (recommended by UK NHS), (2) high and low specificity, (3) high and low sensitivity, and (4) high sensitivity and low specificity readers. For each group, all possible AB double reading pairs were simulated using the OR recall rule. We compared sensitivities and specificities of these complementary pairs first to those of homogeneous (AA, BB) pairs, and then to each other. Statistical significance was determined using Welch’s t-test and 95% confidence intervals. To weigh sensitivity and specificity benefits, ROC curves, Youden’s indices, and positive likelihood ratios were compared.
Result: Grouping according to sensitivity and according to specificity significantly increased sensitivity by 3.5% (p=0.0009) and 1.7% (p=0.037) respectively, compared to homogeneous pairings, with no significant effects on specificity. Grouping high sensitivity and low specificity readers produced a sensitivity of 0.918 (95%CI: 0.913, 0.924), significantly higher than all other groups. Grouping according to experience produced a sensitivity of 0.852 (95%CI: 0.844, 0.859), significantly lower than all other groups, but also the significantly highest specificity, 0.722 (95%CI: 0.707, 0.737). The bootstrap method for ROC comparisons applied to pAUC (sensitivity >0.8) shows that Group (1)’s most extreme readers have the highest performance (non-significant). However, Youden’s indices and positive likelihood ratios show significant differences between the complementary groups with Group (1) still being the highest, followed by Group (3).
Conclusion: Some forms of pairing by complementary ability levels can significantly improve sensitivity, with an insignificant effect on specificity, compared to homogeneous pairings. These preliminary results suggest that pairing by sensitivity yields the best clinical performance, and should be further investigated. Pairing readers simply according to convenience could be significantly less effective.
Since empirical utility scores are now available for all health states related to cervical cancer screening and treatment, we estimated the impact of using these utility scores on the cost-effectiveness of cervical cancer screening, compared to using utility scores from the literature.
We first reviewed the literature on cost-effectiveness analyses (CEAs) of cervical cancer screening published between 2003 and 2013. We focused on studies that used quality adjusted life years (QALYs). We evaluated the differences in utility assumptions between the publications. For the different CEAs and based on the empirical data, we calculated the number of days lost due to loss of quality of life for the different health states, by multiplying the assumed utilities with the mean durations of the loss in quality of life.
We used the microsimulation screening analysis (MISCAN) model to estimate the impact of using these different utility scores on the cost-effectiveness (costs per QALY gained) of primary human papillomavirus (HPV) screening compared to no screening.
Utility scores and number of days lost due to loss of quality of life as assumed for women in different health states that are related to cervical cancer and its prevention are very heterogeneous across the different CEAs, as well as compared to the empirical data. These differences result in a significant variation in cost-effectiveness of primary HPV screening compared to no screening, ranging from €8,035 to €13,518 per QALY gained (See Figure). If the empirical data was used, the cost effectiveness of primary HPV screening was €11,839 per QALY gained.
The assumed number of days lost as a consequence
of loss in quality of life for different health states in CEAs of cervical
cancer screening has a major effect on the estimated cost-effectiveness ratio
of screening. We showed that most CEAs, compared to the empirical data,
overestimated the number of QALYs gained by screening, and therefore
overestimate the cost effectiveness of screening. Utility assessment in CEAs
therefore needs to be based on good quality data. From our analysis, we can
conclude that for comparability, extensive sensitivity analyses on quality of
life assumptions and presentation of costs per life year gained in CEAs are
Method: We start by modelling the natural progression of breast cancer using a Markov process. A patient follows from healthy state to preclinical state and finally to a clinical state, if the cancer is not detected by screening. We incorporate the effects of covariates on transition rates using proportional hazards model. Patients are women aged 50-59 from the Canadian National Breast Screening Study-2 (CNBSS). We include prevalent cancers, which are cancers detected at the initial screen of the CNBSS screening program by means of different modalities. We estimate the screening sensitivities at the initial and subsequent screens for both study and control groups of the CNBSS. We then develop a simulation model to predict the expected number of prevalent, screen-detected and clinical cancers for this age group, and validate the progression model and screening sensitivities.
Result: We compare the effectiveness of mammography (MA) and physical breast examination (PEX) for the study group to that of solely PEX for the control group. The values of the program sensitivity at initial and subsequent screens are given in table 1. Our analysis shows that age of entry and family history are the two main risks factors for cancer progression for the age group 50-59. We estimate that only 83.5% of reported cancer cases would have developed within the follow-up period of five years of the CNBSS if those women had been screened only through PEX. We evaluate overdiagnosis due to the addition of MA screening at 21.8%.
Conclusion: The values we estimated and validated for both control and study groups for transition rate, risk factors, effect of screening modalities and programs and proportion of overdiagnosis will help policy-makers by giving them the tools to improve the QALY and cost of their screening strategies.
|Control group (PEX only)||Study group (MA & PEX)|
|Sensitivity at initial screen||69%||95%|
|Sensitivity at subsequent screens||39%||82%|